kakinang INSIGHTS · 2026-08-28
AI and human touch: put speed, judgment, and responsibility in the right place
Human– AI Help me evaluate: AI and human touch: put speed, judgment, and responsibility in the right place. service quality depends on division of work and handoff, not the label on a reply. Customers need to know who they are speaking with and reach a responsible next person when needed.
Give AI repeatable, verifiable work
Opening hours, public service descriptions, and basic steps can use approved material; unknown or changing facts need a query or an honest limitation.
Implementation example: Let AI answer hours, directions, and published service steps first; route price exceptions or bespoke commitments to an authorised person Assign an accountable owner and align the public page, conversation state, and back-office task; verify both a normal enquiry and missing-data or duplicate-submit cases. The model must not cross confirmation, permission, or staffing boundaries. Review completion, repeat questions, and human corrections weekly, then update the operating rule.
Give judgment and commitments to people
Refunds, exceptions, legal risk, emotional complaints, and special commitments require an accountable person with authority.
Implementation example: For a refund or complaint, create an explicit owner, handling target, and link back to the original conversation—not only a handoff button Assign an accountable owner and align the public page, conversation state, and back-office task; verify both a normal enquiry and missing-data or duplicate-submit cases. The model must not cross confirmation, permission, or staffing boundaries. Review completion, repeat questions, and human corrections weekly, then update the operating rule.
Show identity and mode clearly
- Do not package automation as a human being live. Async, live, and disabled handoff modes should be presented truthfully.
- Implementation example: State plainly whether the surface is AI, asynchronous messaging, or live human support; sending a notice is not a human reply Assign an accountable owner and align the public page, conversation state, and back-office task; verify both a normal enquiry and missing-data or duplicate-submit cases. The model must not cross confirmation, permission, or staffing boundaries. Review completion, repeat questions, and human corrections weekly, then update the operating rule.
Make handoff avoid repetition
Include the question summary, sources used, and consented details when handing to a person, with a way back to the original conversation.
Implementation example: Let staff see sources cited, consented details, and open questions in the summary rather than a model-invented personality label Assign an accountable owner and align the public page, conversation state, and back-office task; verify both a normal enquiry and missing-data or duplicate-submit cases. The model must not cross confirmation, permission, or staffing boundaries. Review completion, repeat questions, and human corrections weekly, then update the operating rule.
Respect a refusal and stop
Do not repeatedly request details after a customer declines or asks to stop. When AI cannot help, offer a courteous alternative path.
Implementation example: When a customer declines a phone number, offer email, a form, or a close and remember the refusal for the next turn Assign an accountable owner and align the public page, conversation state, and back-office task; verify both a normal enquiry and missing-data or duplicate-submit cases. The model must not cross confirmation, permission, or staffing boundaries. Review completion, repeat questions, and human corrections weekly, then update the operating rule.
Turn quality into improvable signals
- Review repeat questions, human corrections, unresolved reasons, and first response instead of treating blocked conversations as the only success metric.
- Implementation example: Sample emotional, repeatedly unresolved, and human-corrected cases weekly to confirm AI does not interrupt after handoff or promise an exception Assign an accountable owner and align the public page, conversation state, and back-office task; verify both a normal enquiry and missing-data or duplicate-submit cases. The model must not cross confirmation, permission, or staffing boundaries. Review completion, repeat questions, and human corrections weekly, then update the operating rule.
Questions teams ask
Will AI make service cold?
Not necessarily. Dishonesty, irrelevant answers, and an unreachable handoff are what undermine trust.
When is human handoff required?
For responsibility, exceptions, strong dissatisfaction, repeated failure, or an explicit customer request.
What if nobody is online?
State the asynchronous path or another contact route clearly; do not promise an immediate reply.
Can AI retain a brand voice?
Yes, within approved service boundaries and copy, but it cannot invent policy or commitments.